1/*
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
39 * $FreeBSD: head/sys/kern/vfs_vnops.c 99009 2002-06-29 01:50:25Z alfred $
40 */
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/fcntl.h>
45#include <sys/file.h>
46#include <sys/stat.h>
47#include <sys/proc.h>
48#include <sys/lock.h>
49#include <sys/mount.h>
50#include <sys/mutex.h>
51#include <sys/namei.h>
52#include <sys/vnode.h>
53#include <sys/bio.h>
54#include <sys/buf.h>
55#include <sys/filio.h>
56#include <sys/sx.h>
57#include <sys/ttycom.h>
58#include <sys/conf.h>
59#include <sys/syslog.h>
60
61#include <machine/limits.h>
62
63static int vn_closefile(struct file *fp, struct thread *td);
64static int vn_ioctl(struct file *fp, u_long com, void *data,
65 struct thread *td);
66static int vn_read(struct file *fp, struct uio *uio,
67 struct ucred *cred, int flags, struct thread *td);
68static int vn_poll(struct file *fp, int events, struct ucred *cred,
69 struct thread *td);
70static int vn_kqfilter(struct file *fp, struct knote *kn);
71static int vn_statfile(struct file *fp, struct stat *sb, struct thread *td);
72static int vn_write(struct file *fp, struct uio *uio,
73 struct ucred *cred, int flags, struct thread *td);
74
75struct fileops vnops = {
76 vn_read, vn_write, vn_ioctl, vn_poll, vn_kqfilter,
77 vn_statfile, vn_closefile
78};
79
80int
81vn_open(ndp, flagp, cmode)
82 register struct nameidata *ndp;
83 int *flagp, cmode;
84{
85 struct thread *td = ndp->ni_cnd.cn_thread;
86
87 return (vn_open_cred(ndp, flagp, cmode, td->td_ucred));
88}
89
90/*
91 * Common code for vnode open operations.
92 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
93 *
94 * Note that this does NOT free nameidata for the successful case,
95 * due to the NDINIT being done elsewhere.
96 */
97int
98vn_open_cred(ndp, flagp, cmode, cred)
99 register struct nameidata *ndp;
100 int *flagp, cmode;
101 struct ucred *cred;
102{
103 struct vnode *vp;
104 struct mount *mp;
105 struct thread *td = ndp->ni_cnd.cn_thread;
106 struct vattr vat;
107 struct vattr *vap = &vat;
108 int mode, fmode, error;
109#ifdef LOOKUP_SHARED
110 int exclusive; /* The current intended lock state */
111
112 exclusive = 0;
113#endif
114
115restart:
116 fmode = *flagp;
117 if (fmode & O_CREAT) {
118 ndp->ni_cnd.cn_nameiop = CREATE;
119 ndp->ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
120 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
121 ndp->ni_cnd.cn_flags |= FOLLOW;
122 bwillwrite();
123 if ((error = namei(ndp)) != 0)
124 return (error);
125 if (ndp->ni_vp == NULL) {
126 VATTR_NULL(vap);
127 vap->va_type = VREG;
128 vap->va_mode = cmode;
129 if (fmode & O_EXCL)
130 vap->va_vaflags |= VA_EXCLUSIVE;
131 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
132 NDFREE(ndp, NDF_ONLY_PNBUF);
133 vput(ndp->ni_dvp);
134 if ((error = vn_start_write(NULL, &mp,
135 V_XSLEEP | PCATCH)) != 0)
136 return (error);
137 goto restart;
138 }
139 VOP_LEASE(ndp->ni_dvp, td, cred, LEASE_WRITE);
140 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
141 &ndp->ni_cnd, vap);
142 vput(ndp->ni_dvp);
143 vn_finished_write(mp);
144 if (error) {
145 NDFREE(ndp, NDF_ONLY_PNBUF);
146 return (error);
147 }
148 ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "create");
149 ASSERT_VOP_LOCKED(ndp->ni_vp, "create");
150 fmode &= ~O_TRUNC;
151 vp = ndp->ni_vp;
152#ifdef LOOKUP_SHARED
153 exclusive = 1;
154#endif
155 } else {
156 if (ndp->ni_dvp == ndp->ni_vp)
157 vrele(ndp->ni_dvp);
158 else
159 vput(ndp->ni_dvp);
160 ndp->ni_dvp = NULL;
161 vp = ndp->ni_vp;
162 if (fmode & O_EXCL) {
163 error = EEXIST;
164 goto bad;
165 }
166 fmode &= ~O_CREAT;
167 }
168 } else {
169 ndp->ni_cnd.cn_nameiop = LOOKUP;
170#ifdef LOOKUP_SHARED
171 ndp->ni_cnd.cn_flags =
172 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
173 LOCKSHARED | LOCKLEAF;
174#else
175 ndp->ni_cnd.cn_flags =
176 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
177#endif
178 if ((error = namei(ndp)) != 0)
179 return (error);
180 vp = ndp->ni_vp;
181 }
182 if (vp->v_type == VLNK) {
183 error = EMLINK;
184 goto bad;
185 }
186 if (vp->v_type == VSOCK) {
187 error = EOPNOTSUPP;
188 goto bad;
189 }
190 if ((fmode & O_CREAT) == 0) {
191 mode = 0;
192 if (fmode & (FWRITE | O_TRUNC)) {
193 if (vp->v_type == VDIR) {
194 error = EISDIR;
195 goto bad;
196 }
197 error = vn_writechk(vp);
198 if (error)
199 goto bad;
200 mode |= VWRITE;
201 }
202 if (fmode & FREAD)
203 mode |= VREAD;
204 if (mode) {
205 error = VOP_ACCESS(vp, mode, cred, td);
206 if (error)
207 goto bad;
208 }
209 }
210 if ((error = VOP_OPEN(vp, fmode, cred, td)) != 0)
211 goto bad;
212 /*
213 * Make sure that a VM object is created for VMIO support.
214 */
215 if (vn_canvmio(vp) == TRUE) {
216#ifdef LOOKUP_SHARED
217 int flock;
218
219 if (!exclusive && VOP_GETVOBJECT(vp, NULL) != 0)
220 VOP_LOCK(vp, LK_UPGRADE, td);
221 /*
222 * In cases where the object is marked as dead object_create
223 * will unlock and relock exclusive. It is safe to call in
224 * here with a shared lock because we only examine fields that
225 * the shared lock guarantees will be stable. In the UPGRADE
226 * case it is not likely that anyone has used this vnode yet
227 * so there will be no contention. The logic after this call
228 * restores the requested locking state.
229 */
230#endif
231 if ((error = vfs_object_create(vp, td, cred)) != 0) {
232 VOP_UNLOCK(vp, 0, td);
233 VOP_CLOSE(vp, fmode, cred, td);
234 NDFREE(ndp, NDF_ONLY_PNBUF);
235 vrele(vp);
236 *flagp = fmode;
237 return (error);
238 }
239#ifdef LOOKUP_SHARED
240 flock = VOP_ISLOCKED(vp, td);
241 if (!exclusive && flock == LK_EXCLUSIVE)
242 VOP_LOCK(vp, LK_DOWNGRADE, td);
243#endif
244 }
245
246 if (fmode & FWRITE)
247 vp->v_writecount++;
248 *flagp = fmode;
249 return (0);
250bad:
251 NDFREE(ndp, NDF_ONLY_PNBUF);
252 vput(vp);
253 *flagp = fmode;
254 return (error);
255}
256
257/*
258 * Check for write permissions on the specified vnode.
259 * Prototype text segments cannot be written.
260 */
261int
262vn_writechk(vp)
263 register struct vnode *vp;
264{
265
266 /*
267 * If there's shared text associated with
268 * the vnode, try to free it up once. If
269 * we fail, we can't allow writing.
270 */
271 if (vp->v_flag & VTEXT)
272 return (ETXTBSY);
273 return (0);
274}
275
276/*
277 * Vnode close call
278 */
279int
280vn_close(vp, flags, cred, td)
281 register struct vnode *vp;
282 int flags;
283 struct ucred *cred;
284 struct thread *td;
285{
286 int error;
287
288 if (flags & FWRITE)
289 vp->v_writecount--;
290 error = VOP_CLOSE(vp, flags, cred, td);
291 /*
292 * XXX - In certain instances VOP_CLOSE has to do the vrele
293 * itself. If the vrele has been done, it will return EAGAIN
294 * to indicate that the vrele should not be done again. When
295 * this happens, we just return success. The correct thing to
296 * do would be to have all VOP_CLOSE instances do the vrele.
297 */
298 if (error == EAGAIN)
299 return (0);
300 vrele(vp);
301 return (error);
302}
303
304/*
305 * Sequential heuristic - detect sequential operation
306 */
307static __inline
308int
309sequential_heuristic(struct uio *uio, struct file *fp)
310{
311
312 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
313 uio->uio_offset == fp->f_nextoff) {
314 /*
315 * XXX we assume that the filesystem block size is
316 * the default. Not true, but still gives us a pretty
317 * good indicator of how sequential the read operations
318 * are.
319 */
320 fp->f_seqcount += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
321 if (fp->f_seqcount >= 127)
322 fp->f_seqcount = 127;
323 return(fp->f_seqcount << 16);
324 }
325
326 /*
327 * Not sequential, quick draw-down of seqcount
328 */
329 if (fp->f_seqcount > 1)
330 fp->f_seqcount = 1;
331 else
332 fp->f_seqcount = 0;
333 return(0);
334}
335
336/*
337 * Package up an I/O request on a vnode into a uio and do it.
338 */
339int
340vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td)
341 enum uio_rw rw;
342 struct vnode *vp;
343 caddr_t base;
344 int len;
345 off_t offset;
346 enum uio_seg segflg;
347 int ioflg;
348 struct ucred *cred;
349 int *aresid;
350 struct thread *td;
351{
352 struct uio auio;
353 struct iovec aiov;
354 struct mount *mp;
355 int error;
356
357 if ((ioflg & IO_NODELOCKED) == 0) {
358 mp = NULL;
359 if (rw == UIO_WRITE) {
360 if (vp->v_type != VCHR &&
361 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
362 != 0)
363 return (error);
364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
365 } else {
366 vn_lock(vp, LK_SHARED | LK_RETRY, td);
367 }
368
369 }
370 auio.uio_iov = &aiov;
371 auio.uio_iovcnt = 1;
372 aiov.iov_base = base;
373 aiov.iov_len = len;
374 auio.uio_resid = len;
375 auio.uio_offset = offset;
376 auio.uio_segflg = segflg;
377 auio.uio_rw = rw;
378 auio.uio_td = td;
379 if (rw == UIO_READ) {
380 error = VOP_READ(vp, &auio, ioflg, cred);
381 } else {
382 error = VOP_WRITE(vp, &auio, ioflg, cred);
383 }
384 if (aresid)
385 *aresid = auio.uio_resid;
386 else
387 if (auio.uio_resid && error == 0)
388 error = EIO;
389 if ((ioflg & IO_NODELOCKED) == 0) {
390 if (rw == UIO_WRITE)
391 vn_finished_write(mp);
392 VOP_UNLOCK(vp, 0, td);
393 }
394 return (error);
395}
396
397/*
398 * Package up an I/O request on a vnode into a uio and do it. The I/O
399 * request is split up into smaller chunks and we try to avoid saturating
400 * the buffer cache while potentially holding a vnode locked, so we
401 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield()
402 * to give other processes a chance to lock the vnode (either other processes
403 * core'ing the same binary, or unrelated processes scanning the directory).
404 */
405int
406vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td)
407 enum uio_rw rw;
408 struct vnode *vp;
409 caddr_t base;
410 int len;
411 off_t offset;
412 enum uio_seg segflg;
413 int ioflg;
414 struct ucred *cred;
415 int *aresid;
416 struct thread *td;
417{
418 int error = 0;
419
420 do {
421 int chunk = (len > MAXBSIZE) ? MAXBSIZE : len;
422
423 if (rw != UIO_READ && vp->v_type == VREG)
424 bwillwrite();
425 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
426 ioflg, cred, aresid, td);
427 len -= chunk; /* aresid calc already includes length */
428 if (error)
429 break;
430 offset += chunk;
431 base += chunk;
432 uio_yield();
433 } while (len);
434 if (aresid)
435 *aresid += len;
436 return (error);
437}
438
439/*
440 * File table vnode read routine.
441 */
442static int
443vn_read(fp, uio, cred, flags, td)
444 struct file *fp;
445 struct uio *uio;
446 struct ucred *cred;
447 struct thread *td;
448 int flags;
449{
450 struct vnode *vp;
451 int error, ioflag;
452
453 mtx_lock(&Giant);
454 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
455 uio->uio_td, td));
456 vp = (struct vnode *)fp->f_data;
457 ioflag = 0;
458 if (fp->f_flag & FNONBLOCK)
459 ioflag |= IO_NDELAY;
460 if (fp->f_flag & O_DIRECT)
461 ioflag |= IO_DIRECT;
462 VOP_LEASE(vp, td, cred, LEASE_READ);
463 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
464 if ((flags & FOF_OFFSET) == 0)
465 uio->uio_offset = fp->f_offset;
466
467 ioflag |= sequential_heuristic(uio, fp);
468
469 error = VOP_READ(vp, uio, ioflag, cred);
470 if ((flags & FOF_OFFSET) == 0)
471 fp->f_offset = uio->uio_offset;
472 fp->f_nextoff = uio->uio_offset;
473 VOP_UNLOCK(vp, 0, td);
474 mtx_unlock(&Giant);
475 return (error);
476}
477
478/*
479 * File table vnode write routine.
480 */
481static int
482vn_write(fp, uio, cred, flags, td)
483 struct file *fp;
484 struct uio *uio;
485 struct ucred *cred;
486 struct thread *td;
487 int flags;
488{
489 struct vnode *vp;
490 struct mount *mp;
491 int error, ioflag;
492
493 mtx_lock(&Giant);
494 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
495 uio->uio_td, td));
496 vp = (struct vnode *)fp->f_data;
497 if (vp->v_type == VREG)
498 bwillwrite();
499 ioflag = IO_UNIT;
500 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
501 ioflag |= IO_APPEND;
502 if (fp->f_flag & FNONBLOCK)
503 ioflag |= IO_NDELAY;
504 if (fp->f_flag & O_DIRECT)
505 ioflag |= IO_DIRECT;
506 if ((fp->f_flag & O_FSYNC) ||
507 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
508 ioflag |= IO_SYNC;
509 mp = NULL;
510 if (vp->v_type != VCHR &&
511 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) {
512 mtx_unlock(&Giant);
513 return (error);
514 }
515 VOP_LEASE(vp, td, cred, LEASE_WRITE);
516 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
517 if ((flags & FOF_OFFSET) == 0)
518 uio->uio_offset = fp->f_offset;
519 ioflag |= sequential_heuristic(uio, fp);
520 error = VOP_WRITE(vp, uio, ioflag, cred);
521 if ((flags & FOF_OFFSET) == 0)
522 fp->f_offset = uio->uio_offset;
523 fp->f_nextoff = uio->uio_offset;
524 VOP_UNLOCK(vp, 0, td);
525 vn_finished_write(mp);
526 mtx_unlock(&Giant);
527 return (error);
528}
529
530/*
531 * File table vnode stat routine.
532 */
533static int
534vn_statfile(fp, sb, td)
535 struct file *fp;
536 struct stat *sb;
537 struct thread *td;
538{
539 struct vnode *vp = (struct vnode *)fp->f_data;
540 int error;
541
542 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
543 error = vn_stat(vp, sb, td);
544 VOP_UNLOCK(vp, 0, td);
545
546 return (error);
547}
548
549/*
550 * Stat a vnode; implementation for the stat syscall
551 */
552int
553vn_stat(vp, sb, td)
554 struct vnode *vp;
555 register struct stat *sb;
556 struct thread *td;
557{
558 struct vattr vattr;
559 register struct vattr *vap;
560 int error;
561 u_short mode;
562
563 vap = &vattr;
564 error = VOP_GETATTR(vp, vap, td->td_ucred, td);
565 if (error)
566 return (error);
567
568 /*
569 * Zero the spare stat fields
570 */
571 bzero(sb, sizeof *sb);
572
573 /*
574 * Copy from vattr table
575 */
576 if (vap->va_fsid != VNOVAL)
577 sb->st_dev = vap->va_fsid;
578 else
579 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
580 sb->st_ino = vap->va_fileid;
581 mode = vap->va_mode;
582 switch (vap->va_type) {
583 case VREG:
584 mode |= S_IFREG;
585 break;
586 case VDIR:
587 mode |= S_IFDIR;
588 break;
589 case VBLK:
590 mode |= S_IFBLK;
591 break;
592 case VCHR:
593 mode |= S_IFCHR;
594 break;
595 case VLNK:
596 mode |= S_IFLNK;
597 /* This is a cosmetic change, symlinks do not have a mode. */
598 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
599 sb->st_mode &= ~ACCESSPERMS; /* 0000 */
600 else
601 sb->st_mode |= ACCESSPERMS; /* 0777 */
602 break;
603 case VSOCK:
604 mode |= S_IFSOCK;
605 break;
606 case VFIFO:
607 mode |= S_IFIFO;
608 break;
609 default:
610 return (EBADF);
611 };
612 sb->st_mode = mode;
613 sb->st_nlink = vap->va_nlink;
614 sb->st_uid = vap->va_uid;
615 sb->st_gid = vap->va_gid;
616 sb->st_rdev = vap->va_rdev;
617 if (vap->va_size > OFF_MAX)
618 return (EOVERFLOW);
619 sb->st_size = vap->va_size;
620 sb->st_atimespec = vap->va_atime;
621 sb->st_mtimespec = vap->va_mtime;
622 sb->st_ctimespec = vap->va_ctime;
623 sb->st_createtimespec = vap->va_createtime;
624
625 /*
626 * According to www.opengroup.org, the meaning of st_blksize is
627 * "a filesystem-specific preferred I/O block size for this
628 * object. In some filesystem types, this may vary from file
629 * to file"
630 * Default to PAGE_SIZE after much discussion.
631 */
632
633 if (vap->va_type == VREG) {
634 sb->st_blksize = vap->va_blocksize;
635 } else if (vn_isdisk(vp, NULL)) {
636 sb->st_blksize = vp->v_rdev->si_bsize_best;
637 if (sb->st_blksize < vp->v_rdev->si_bsize_phys)
638 sb->st_blksize = vp->v_rdev->si_bsize_phys;
639 if (sb->st_blksize < BLKDEV_IOSIZE)
640 sb->st_blksize = BLKDEV_IOSIZE;
641 } else {
642 sb->st_blksize = PAGE_SIZE;
643 }
644
645 sb->st_flags = vap->va_flags;
646 if (suser(td))
647 sb->st_gen = 0;
648 else
649 sb->st_gen = vap->va_gen;
650
651#if (S_BLKSIZE == 512)
652 /* Optimize this case */
653 sb->st_blocks = vap->va_bytes >> 9;
654#else
655 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
656#endif
657 return (0);
658}
659
660/*
661 * File table vnode ioctl routine.
662 */
663static int
664vn_ioctl(fp, com, data, td)
665 struct file *fp;
666 u_long com;
667 void *data;
668 struct thread *td;
669{
670 register struct vnode *vp = ((struct vnode *)fp->f_data);
671 struct vnode *vpold;
672 struct vattr vattr;
673 int error;
674
675 switch (vp->v_type) {
676
677 case VREG:
678 case VDIR:
679 if (com == FIONREAD) {
680 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
681 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
682 VOP_UNLOCK(vp, 0, td);
683 if (error)
684 return (error);
685 *(int *)data = vattr.va_size - fp->f_offset;
686 return (0);
687 }
688 if (com == FIONBIO || com == FIOASYNC) /* XXX */
689 return (0); /* XXX */
690 /* fall into ... */
691
692 default:
693#if 0
694 return (ENOTTY);
695#endif
696 case VFIFO:
697 case VCHR:
698 case VBLK:
699 if (com == FIODTYPE) {
700 if (vp->v_type != VCHR && vp->v_type != VBLK)
701 return (ENOTTY);
702 *(int *)data = devsw(vp->v_rdev)->d_flags & D_TYPEMASK;
703 return (0);
704 }
705 error = VOP_IOCTL(vp, com, data, fp->f_flag, td->td_ucred, td);
706 if (error == 0 && com == TIOCSCTTY) {
707
708 /* Do nothing if reassigning same control tty */
709 sx_slock(&proctree_lock);
710 if (td->td_proc->p_session->s_ttyvp == vp) {
711 sx_sunlock(&proctree_lock);
712 return (0);
713 }
714
715 vpold = td->td_proc->p_session->s_ttyvp;
716 VREF(vp);
717 SESS_LOCK(td->td_proc->p_session);
718 td->td_proc->p_session->s_ttyvp = vp;
719 SESS_UNLOCK(td->td_proc->p_session);
720
721 sx_sunlock(&proctree_lock);
722
723 /* Get rid of reference to old control tty */
724 if (vpold)
725 vrele(vpold);
726 }
727 return (error);
728 }
729}
730
731/*
732 * File table vnode poll routine.
733 */
734static int
735vn_poll(fp, events, cred, td)
736 struct file *fp;
737 int events;
738 struct ucred *cred;
739 struct thread *td;
740{
741
742 return (VOP_POLL(((struct vnode *)fp->f_data), events, cred, td));
743}
744
745/*
746 * Check that the vnode is still valid, and if so
747 * acquire requested lock.
748 */
749int
750#ifndef DEBUG_LOCKS
751vn_lock(vp, flags, td)
752#else
753debug_vn_lock(vp, flags, td, filename, line)
754#endif
755 struct vnode *vp;
756 int flags;
757 struct thread *td;
758#ifdef DEBUG_LOCKS
759 const char *filename;
760 int line;
761#endif
762{
763 int error;
764
765 do {
766 if ((flags & LK_INTERLOCK) == 0)
767 mtx_lock(&vp->v_interlock);
768 if ((vp->v_flag & VXLOCK) && vp->v_vxproc != curthread) {
769 vp->v_flag |= VXWANT;
770 msleep(vp, &vp->v_interlock, PINOD | PDROP,
771 "vn_lock", 0);
772 error = ENOENT;
773 } else {
774#if 0
775 /* this can now occur in normal operation */
776 if (vp->v_vxproc != NULL)
777 log(LOG_INFO, "VXLOCK interlock avoided in vn_lock\n");
778#endif
779#ifdef DEBUG_LOCKS
780 vp->filename = filename;
781 vp->line = line;
782#endif
783 error = VOP_LOCK(vp,
784 flags | LK_NOPAUSE | LK_INTERLOCK, td);
785 if (error == 0)
786 return (error);
787 }
788 flags &= ~LK_INTERLOCK;
789 } while (flags & LK_RETRY);
790 return (error);
791}
792
793/*
794 * File table vnode close routine.
795 */
796static int
797vn_closefile(fp, td)
798 struct file *fp;
799 struct thread *td;
800{
801
802 fp->f_ops = &badfileops;
803 return (vn_close(((struct vnode *)fp->f_data), fp->f_flag,
804 fp->f_cred, td));
805}
806
807/*
808 * Preparing to start a filesystem write operation. If the operation is
809 * permitted, then we bump the count of operations in progress and
810 * proceed. If a suspend request is in progress, we wait until the
811 * suspension is over, and then proceed.
812 */
813int
814vn_start_write(vp, mpp, flags)
815 struct vnode *vp;
816 struct mount **mpp;
817 int flags;
818{
819 struct mount *mp;
820 int error;
821
822 /*
823 * If a vnode is provided, get and return the mount point that
824 * to which it will write.
825 */
826 if (vp != NULL) {
827 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
828 *mpp = NULL;
829 if (error != EOPNOTSUPP)
830 return (error);
831 return (0);
832 }
833 }
834 if ((mp = *mpp) == NULL)
835 return (0);
836 /*
837 * Check on status of suspension.
838 */
839 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
840 if (flags & V_NOWAIT)
841 return (EWOULDBLOCK);
842 error = tsleep(&mp->mnt_flag, (PUSER - 1) | (flags & PCATCH),
843 "suspfs", 0);
844 if (error)
845 return (error);
846 }
847 if (flags & V_XSLEEP)
848 return (0);
849 mp->mnt_writeopcount++;
850 return (0);
851}
852
853/*
854 * Secondary suspension. Used by operations such as vop_inactive
855 * routines that are needed by the higher level functions. These
856 * are allowed to proceed until all the higher level functions have
857 * completed (indicated by mnt_writeopcount dropping to zero). At that
858 * time, these operations are halted until the suspension is over.
859 */
860int
861vn_write_suspend_wait(vp, mp, flags)
862 struct vnode *vp;
863 struct mount *mp;
864 int flags;
865{
866 int error;
867
868 if (vp != NULL) {
869 if ((error = VOP_GETWRITEMOUNT(vp, &mp)) != 0) {
870 if (error != EOPNOTSUPP)
871 return (error);
872 return (0);
873 }
874 }
875 /*
876 * If we are not suspended or have not yet reached suspended
877 * mode, then let the operation proceed.
878 */
879 if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPENDED) == 0)
880 return (0);
881 if (flags & V_NOWAIT)
882 return (EWOULDBLOCK);
883 /*
884 * Wait for the suspension to finish.
885 */
886 return (tsleep(&mp->mnt_flag, (PUSER - 1) | (flags & PCATCH),
887 "suspfs", 0));
888}
889
890/*
891 * Filesystem write operation has completed. If we are suspending and this
892 * operation is the last one, notify the suspender that the suspension is
893 * now in effect.
894 */
895void
896vn_finished_write(mp)
897 struct mount *mp;
898{
899
900 if (mp == NULL)
901 return;
902 mp->mnt_writeopcount--;
903 if (mp->mnt_writeopcount < 0)
904 panic("vn_finished_write: neg cnt");
905 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
906 mp->mnt_writeopcount <= 0)
907 wakeup(&mp->mnt_writeopcount);
908}
909
910/*
911 * Request a filesystem to suspend write operations.
912 */
913void
914vfs_write_suspend(mp)
915 struct mount *mp;
916{
917 struct thread *td = curthread;
918
919 if (mp->mnt_kern_flag & MNTK_SUSPEND)
920 return;
921 mp->mnt_kern_flag |= MNTK_SUSPEND;
922 if (mp->mnt_writeopcount > 0)
923 (void) tsleep(&mp->mnt_writeopcount, PUSER - 1, "suspwt", 0);
924 VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td);
925 mp->mnt_kern_flag |= MNTK_SUSPENDED;
926}
927
928/*
929 * Request a filesystem to resume write operations.
930 */
931void
932vfs_write_resume(mp)
933 struct mount *mp;
934{
935
936 if ((mp->mnt_kern_flag & MNTK_SUSPEND) == 0)
937 return;
938 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPENDED);
939 wakeup(&mp->mnt_writeopcount);
940 wakeup(&mp->mnt_flag);
941}
942
943/*
944 * Implement kqueues for files by translating it to vnode operation.
945 */
946static int
947vn_kqfilter(struct file *fp, struct knote *kn)
948{
949
950 return (VOP_KQFILTER(((struct vnode *)fp->f_data), kn));
951}
952
953/*
954 * Simplified in-kernel wrapper calls for extended attribute access.
955 * Both calls pass in a NULL credential, authorizing as "kernel" access.
956 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
957 */
958int
959vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
960 const char *attrname, int *buflen, char *buf, struct thread *td)
961{
962 struct uio auio;
963 struct iovec iov;
964 int error;
965
966 iov.iov_len = *buflen;
967 iov.iov_base = buf;
968
969 auio.uio_iov = &iov;
970 auio.uio_iovcnt = 1;
971 auio.uio_rw = UIO_READ;
972 auio.uio_segflg = UIO_SYSSPACE;
973 auio.uio_td = td;
974 auio.uio_offset = 0;
975 auio.uio_resid = *buflen;
976
977 if ((ioflg & IO_NODELOCKED) == 0)
978 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
979
980 /* authorize attribute retrieval as kernel */
981 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
982 td);
983
984 if ((ioflg & IO_NODELOCKED) == 0)
985 VOP_UNLOCK(vp, 0, td);
986
987 if (error == 0) {
988 *buflen = *buflen - auio.uio_resid;
989 }
990
991 return (error);
992}
993
994/*
995 * XXX failure mode if partially written?
996 */
997int
998vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
999 const char *attrname, int buflen, char *buf, struct thread *td)
1000{
1001 struct uio auio;
1002 struct iovec iov;
1003 struct mount *mp;
1004 int error;
1005
1006 iov.iov_len = buflen;
1007 iov.iov_base = buf;
1008
1009 auio.uio_iov = &iov;
1010 auio.uio_iovcnt = 1;
1011 auio.uio_rw = UIO_WRITE;
1012 auio.uio_segflg = UIO_SYSSPACE;
1013 auio.uio_td = td;
1014 auio.uio_offset = 0;
1015 auio.uio_resid = buflen;
1016
1017 if ((ioflg & IO_NODELOCKED) == 0) {
1018 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1019 return (error);
1020 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1021 }
1022
1023 /* authorize attribute setting as kernel */
1024 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1025
1026 if ((ioflg & IO_NODELOCKED) == 0) {
1027 vn_finished_write(mp);
1028 VOP_UNLOCK(vp, 0, td);
1029 }
1030
1031 return (error);
1032}
1033
1034int
1035vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1036 const char *attrname, struct thread *td)
1037{
1038 struct mount *mp;
1039 int error;
1040
1041 if ((ioflg & IO_NODELOCKED) == 0) {
1042 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1043 return (error);
1044 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1045 }
1046
1047 /* authorize attribute removal as kernel */
1048 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td);
1049
1050 if ((ioflg & IO_NODELOCKED) == 0) {
1051 vn_finished_write(mp);
1052 VOP_UNLOCK(vp, 0, td);
1053 }
1054
1055 return (error);
1056}
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
39 * $FreeBSD: head/sys/kern/vfs_vnops.c 99009 2002-06-29 01:50:25Z alfred $
40 */
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/fcntl.h>
45#include <sys/file.h>
46#include <sys/stat.h>
47#include <sys/proc.h>
48#include <sys/lock.h>
49#include <sys/mount.h>
50#include <sys/mutex.h>
51#include <sys/namei.h>
52#include <sys/vnode.h>
53#include <sys/bio.h>
54#include <sys/buf.h>
55#include <sys/filio.h>
56#include <sys/sx.h>
57#include <sys/ttycom.h>
58#include <sys/conf.h>
59#include <sys/syslog.h>
60
61#include <machine/limits.h>
62
63static int vn_closefile(struct file *fp, struct thread *td);
64static int vn_ioctl(struct file *fp, u_long com, void *data,
65 struct thread *td);
66static int vn_read(struct file *fp, struct uio *uio,
67 struct ucred *cred, int flags, struct thread *td);
68static int vn_poll(struct file *fp, int events, struct ucred *cred,
69 struct thread *td);
70static int vn_kqfilter(struct file *fp, struct knote *kn);
71static int vn_statfile(struct file *fp, struct stat *sb, struct thread *td);
72static int vn_write(struct file *fp, struct uio *uio,
73 struct ucred *cred, int flags, struct thread *td);
74
75struct fileops vnops = {
76 vn_read, vn_write, vn_ioctl, vn_poll, vn_kqfilter,
77 vn_statfile, vn_closefile
78};
79
80int
81vn_open(ndp, flagp, cmode)
82 register struct nameidata *ndp;
83 int *flagp, cmode;
84{
85 struct thread *td = ndp->ni_cnd.cn_thread;
86
87 return (vn_open_cred(ndp, flagp, cmode, td->td_ucred));
88}
89
90/*
91 * Common code for vnode open operations.
92 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
93 *
94 * Note that this does NOT free nameidata for the successful case,
95 * due to the NDINIT being done elsewhere.
96 */
97int
98vn_open_cred(ndp, flagp, cmode, cred)
99 register struct nameidata *ndp;
100 int *flagp, cmode;
101 struct ucred *cred;
102{
103 struct vnode *vp;
104 struct mount *mp;
105 struct thread *td = ndp->ni_cnd.cn_thread;
106 struct vattr vat;
107 struct vattr *vap = &vat;
108 int mode, fmode, error;
109#ifdef LOOKUP_SHARED
110 int exclusive; /* The current intended lock state */
111
112 exclusive = 0;
113#endif
114
115restart:
116 fmode = *flagp;
117 if (fmode & O_CREAT) {
118 ndp->ni_cnd.cn_nameiop = CREATE;
119 ndp->ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
120 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
121 ndp->ni_cnd.cn_flags |= FOLLOW;
122 bwillwrite();
123 if ((error = namei(ndp)) != 0)
124 return (error);
125 if (ndp->ni_vp == NULL) {
126 VATTR_NULL(vap);
127 vap->va_type = VREG;
128 vap->va_mode = cmode;
129 if (fmode & O_EXCL)
130 vap->va_vaflags |= VA_EXCLUSIVE;
131 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
132 NDFREE(ndp, NDF_ONLY_PNBUF);
133 vput(ndp->ni_dvp);
134 if ((error = vn_start_write(NULL, &mp,
135 V_XSLEEP | PCATCH)) != 0)
136 return (error);
137 goto restart;
138 }
139 VOP_LEASE(ndp->ni_dvp, td, cred, LEASE_WRITE);
140 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
141 &ndp->ni_cnd, vap);
142 vput(ndp->ni_dvp);
143 vn_finished_write(mp);
144 if (error) {
145 NDFREE(ndp, NDF_ONLY_PNBUF);
146 return (error);
147 }
148 ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "create");
149 ASSERT_VOP_LOCKED(ndp->ni_vp, "create");
150 fmode &= ~O_TRUNC;
151 vp = ndp->ni_vp;
152#ifdef LOOKUP_SHARED
153 exclusive = 1;
154#endif
155 } else {
156 if (ndp->ni_dvp == ndp->ni_vp)
157 vrele(ndp->ni_dvp);
158 else
159 vput(ndp->ni_dvp);
160 ndp->ni_dvp = NULL;
161 vp = ndp->ni_vp;
162 if (fmode & O_EXCL) {
163 error = EEXIST;
164 goto bad;
165 }
166 fmode &= ~O_CREAT;
167 }
168 } else {
169 ndp->ni_cnd.cn_nameiop = LOOKUP;
170#ifdef LOOKUP_SHARED
171 ndp->ni_cnd.cn_flags =
172 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
173 LOCKSHARED | LOCKLEAF;
174#else
175 ndp->ni_cnd.cn_flags =
176 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
177#endif
178 if ((error = namei(ndp)) != 0)
179 return (error);
180 vp = ndp->ni_vp;
181 }
182 if (vp->v_type == VLNK) {
183 error = EMLINK;
184 goto bad;
185 }
186 if (vp->v_type == VSOCK) {
187 error = EOPNOTSUPP;
188 goto bad;
189 }
190 if ((fmode & O_CREAT) == 0) {
191 mode = 0;
192 if (fmode & (FWRITE | O_TRUNC)) {
193 if (vp->v_type == VDIR) {
194 error = EISDIR;
195 goto bad;
196 }
197 error = vn_writechk(vp);
198 if (error)
199 goto bad;
200 mode |= VWRITE;
201 }
202 if (fmode & FREAD)
203 mode |= VREAD;
204 if (mode) {
205 error = VOP_ACCESS(vp, mode, cred, td);
206 if (error)
207 goto bad;
208 }
209 }
210 if ((error = VOP_OPEN(vp, fmode, cred, td)) != 0)
211 goto bad;
212 /*
213 * Make sure that a VM object is created for VMIO support.
214 */
215 if (vn_canvmio(vp) == TRUE) {
216#ifdef LOOKUP_SHARED
217 int flock;
218
219 if (!exclusive && VOP_GETVOBJECT(vp, NULL) != 0)
220 VOP_LOCK(vp, LK_UPGRADE, td);
221 /*
222 * In cases where the object is marked as dead object_create
223 * will unlock and relock exclusive. It is safe to call in
224 * here with a shared lock because we only examine fields that
225 * the shared lock guarantees will be stable. In the UPGRADE
226 * case it is not likely that anyone has used this vnode yet
227 * so there will be no contention. The logic after this call
228 * restores the requested locking state.
229 */
230#endif
231 if ((error = vfs_object_create(vp, td, cred)) != 0) {
232 VOP_UNLOCK(vp, 0, td);
233 VOP_CLOSE(vp, fmode, cred, td);
234 NDFREE(ndp, NDF_ONLY_PNBUF);
235 vrele(vp);
236 *flagp = fmode;
237 return (error);
238 }
239#ifdef LOOKUP_SHARED
240 flock = VOP_ISLOCKED(vp, td);
241 if (!exclusive && flock == LK_EXCLUSIVE)
242 VOP_LOCK(vp, LK_DOWNGRADE, td);
243#endif
244 }
245
246 if (fmode & FWRITE)
247 vp->v_writecount++;
248 *flagp = fmode;
249 return (0);
250bad:
251 NDFREE(ndp, NDF_ONLY_PNBUF);
252 vput(vp);
253 *flagp = fmode;
254 return (error);
255}
256
257/*
258 * Check for write permissions on the specified vnode.
259 * Prototype text segments cannot be written.
260 */
261int
262vn_writechk(vp)
263 register struct vnode *vp;
264{
265
266 /*
267 * If there's shared text associated with
268 * the vnode, try to free it up once. If
269 * we fail, we can't allow writing.
270 */
271 if (vp->v_flag & VTEXT)
272 return (ETXTBSY);
273 return (0);
274}
275
276/*
277 * Vnode close call
278 */
279int
280vn_close(vp, flags, cred, td)
281 register struct vnode *vp;
282 int flags;
283 struct ucred *cred;
284 struct thread *td;
285{
286 int error;
287
288 if (flags & FWRITE)
289 vp->v_writecount--;
290 error = VOP_CLOSE(vp, flags, cred, td);
291 /*
292 * XXX - In certain instances VOP_CLOSE has to do the vrele
293 * itself. If the vrele has been done, it will return EAGAIN
294 * to indicate that the vrele should not be done again. When
295 * this happens, we just return success. The correct thing to
296 * do would be to have all VOP_CLOSE instances do the vrele.
297 */
298 if (error == EAGAIN)
299 return (0);
300 vrele(vp);
301 return (error);
302}
303
304/*
305 * Sequential heuristic - detect sequential operation
306 */
307static __inline
308int
309sequential_heuristic(struct uio *uio, struct file *fp)
310{
311
312 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
313 uio->uio_offset == fp->f_nextoff) {
314 /*
315 * XXX we assume that the filesystem block size is
316 * the default. Not true, but still gives us a pretty
317 * good indicator of how sequential the read operations
318 * are.
319 */
320 fp->f_seqcount += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
321 if (fp->f_seqcount >= 127)
322 fp->f_seqcount = 127;
323 return(fp->f_seqcount << 16);
324 }
325
326 /*
327 * Not sequential, quick draw-down of seqcount
328 */
329 if (fp->f_seqcount > 1)
330 fp->f_seqcount = 1;
331 else
332 fp->f_seqcount = 0;
333 return(0);
334}
335
336/*
337 * Package up an I/O request on a vnode into a uio and do it.
338 */
339int
340vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td)
341 enum uio_rw rw;
342 struct vnode *vp;
343 caddr_t base;
344 int len;
345 off_t offset;
346 enum uio_seg segflg;
347 int ioflg;
348 struct ucred *cred;
349 int *aresid;
350 struct thread *td;
351{
352 struct uio auio;
353 struct iovec aiov;
354 struct mount *mp;
355 int error;
356
357 if ((ioflg & IO_NODELOCKED) == 0) {
358 mp = NULL;
359 if (rw == UIO_WRITE) {
360 if (vp->v_type != VCHR &&
361 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
362 != 0)
363 return (error);
364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
365 } else {
366 vn_lock(vp, LK_SHARED | LK_RETRY, td);
367 }
368
369 }
370 auio.uio_iov = &aiov;
371 auio.uio_iovcnt = 1;
372 aiov.iov_base = base;
373 aiov.iov_len = len;
374 auio.uio_resid = len;
375 auio.uio_offset = offset;
376 auio.uio_segflg = segflg;
377 auio.uio_rw = rw;
378 auio.uio_td = td;
379 if (rw == UIO_READ) {
380 error = VOP_READ(vp, &auio, ioflg, cred);
381 } else {
382 error = VOP_WRITE(vp, &auio, ioflg, cred);
383 }
384 if (aresid)
385 *aresid = auio.uio_resid;
386 else
387 if (auio.uio_resid && error == 0)
388 error = EIO;
389 if ((ioflg & IO_NODELOCKED) == 0) {
390 if (rw == UIO_WRITE)
391 vn_finished_write(mp);
392 VOP_UNLOCK(vp, 0, td);
393 }
394 return (error);
395}
396
397/*
398 * Package up an I/O request on a vnode into a uio and do it. The I/O
399 * request is split up into smaller chunks and we try to avoid saturating
400 * the buffer cache while potentially holding a vnode locked, so we
401 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield()
402 * to give other processes a chance to lock the vnode (either other processes
403 * core'ing the same binary, or unrelated processes scanning the directory).
404 */
405int
406vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td)
407 enum uio_rw rw;
408 struct vnode *vp;
409 caddr_t base;
410 int len;
411 off_t offset;
412 enum uio_seg segflg;
413 int ioflg;
414 struct ucred *cred;
415 int *aresid;
416 struct thread *td;
417{
418 int error = 0;
419
420 do {
421 int chunk = (len > MAXBSIZE) ? MAXBSIZE : len;
422
423 if (rw != UIO_READ && vp->v_type == VREG)
424 bwillwrite();
425 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
426 ioflg, cred, aresid, td);
427 len -= chunk; /* aresid calc already includes length */
428 if (error)
429 break;
430 offset += chunk;
431 base += chunk;
432 uio_yield();
433 } while (len);
434 if (aresid)
435 *aresid += len;
436 return (error);
437}
438
439/*
440 * File table vnode read routine.
441 */
442static int
443vn_read(fp, uio, cred, flags, td)
444 struct file *fp;
445 struct uio *uio;
446 struct ucred *cred;
447 struct thread *td;
448 int flags;
449{
450 struct vnode *vp;
451 int error, ioflag;
452
453 mtx_lock(&Giant);
454 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
455 uio->uio_td, td));
456 vp = (struct vnode *)fp->f_data;
457 ioflag = 0;
458 if (fp->f_flag & FNONBLOCK)
459 ioflag |= IO_NDELAY;
460 if (fp->f_flag & O_DIRECT)
461 ioflag |= IO_DIRECT;
462 VOP_LEASE(vp, td, cred, LEASE_READ);
463 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
464 if ((flags & FOF_OFFSET) == 0)
465 uio->uio_offset = fp->f_offset;
466
467 ioflag |= sequential_heuristic(uio, fp);
468
469 error = VOP_READ(vp, uio, ioflag, cred);
470 if ((flags & FOF_OFFSET) == 0)
471 fp->f_offset = uio->uio_offset;
472 fp->f_nextoff = uio->uio_offset;
473 VOP_UNLOCK(vp, 0, td);
474 mtx_unlock(&Giant);
475 return (error);
476}
477
478/*
479 * File table vnode write routine.
480 */
481static int
482vn_write(fp, uio, cred, flags, td)
483 struct file *fp;
484 struct uio *uio;
485 struct ucred *cred;
486 struct thread *td;
487 int flags;
488{
489 struct vnode *vp;
490 struct mount *mp;
491 int error, ioflag;
492
493 mtx_lock(&Giant);
494 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
495 uio->uio_td, td));
496 vp = (struct vnode *)fp->f_data;
497 if (vp->v_type == VREG)
498 bwillwrite();
499 ioflag = IO_UNIT;
500 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
501 ioflag |= IO_APPEND;
502 if (fp->f_flag & FNONBLOCK)
503 ioflag |= IO_NDELAY;
504 if (fp->f_flag & O_DIRECT)
505 ioflag |= IO_DIRECT;
506 if ((fp->f_flag & O_FSYNC) ||
507 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
508 ioflag |= IO_SYNC;
509 mp = NULL;
510 if (vp->v_type != VCHR &&
511 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) {
512 mtx_unlock(&Giant);
513 return (error);
514 }
515 VOP_LEASE(vp, td, cred, LEASE_WRITE);
516 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
517 if ((flags & FOF_OFFSET) == 0)
518 uio->uio_offset = fp->f_offset;
519 ioflag |= sequential_heuristic(uio, fp);
520 error = VOP_WRITE(vp, uio, ioflag, cred);
521 if ((flags & FOF_OFFSET) == 0)
522 fp->f_offset = uio->uio_offset;
523 fp->f_nextoff = uio->uio_offset;
524 VOP_UNLOCK(vp, 0, td);
525 vn_finished_write(mp);
526 mtx_unlock(&Giant);
527 return (error);
528}
529
530/*
531 * File table vnode stat routine.
532 */
533static int
534vn_statfile(fp, sb, td)
535 struct file *fp;
536 struct stat *sb;
537 struct thread *td;
538{
539 struct vnode *vp = (struct vnode *)fp->f_data;
540 int error;
541
542 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
543 error = vn_stat(vp, sb, td);
544 VOP_UNLOCK(vp, 0, td);
545
546 return (error);
547}
548
549/*
550 * Stat a vnode; implementation for the stat syscall
551 */
552int
553vn_stat(vp, sb, td)
554 struct vnode *vp;
555 register struct stat *sb;
556 struct thread *td;
557{
558 struct vattr vattr;
559 register struct vattr *vap;
560 int error;
561 u_short mode;
562
563 vap = &vattr;
564 error = VOP_GETATTR(vp, vap, td->td_ucred, td);
565 if (error)
566 return (error);
567
568 /*
569 * Zero the spare stat fields
570 */
571 bzero(sb, sizeof *sb);
572
573 /*
574 * Copy from vattr table
575 */
576 if (vap->va_fsid != VNOVAL)
577 sb->st_dev = vap->va_fsid;
578 else
579 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
580 sb->st_ino = vap->va_fileid;
581 mode = vap->va_mode;
582 switch (vap->va_type) {
583 case VREG:
584 mode |= S_IFREG;
585 break;
586 case VDIR:
587 mode |= S_IFDIR;
588 break;
589 case VBLK:
590 mode |= S_IFBLK;
591 break;
592 case VCHR:
593 mode |= S_IFCHR;
594 break;
595 case VLNK:
596 mode |= S_IFLNK;
597 /* This is a cosmetic change, symlinks do not have a mode. */
598 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
599 sb->st_mode &= ~ACCESSPERMS; /* 0000 */
600 else
601 sb->st_mode |= ACCESSPERMS; /* 0777 */
602 break;
603 case VSOCK:
604 mode |= S_IFSOCK;
605 break;
606 case VFIFO:
607 mode |= S_IFIFO;
608 break;
609 default:
610 return (EBADF);
611 };
612 sb->st_mode = mode;
613 sb->st_nlink = vap->va_nlink;
614 sb->st_uid = vap->va_uid;
615 sb->st_gid = vap->va_gid;
616 sb->st_rdev = vap->va_rdev;
617 if (vap->va_size > OFF_MAX)
618 return (EOVERFLOW);
619 sb->st_size = vap->va_size;
620 sb->st_atimespec = vap->va_atime;
621 sb->st_mtimespec = vap->va_mtime;
622 sb->st_ctimespec = vap->va_ctime;
623 sb->st_createtimespec = vap->va_createtime;
624
625 /*
626 * According to www.opengroup.org, the meaning of st_blksize is
627 * "a filesystem-specific preferred I/O block size for this
628 * object. In some filesystem types, this may vary from file
629 * to file"
630 * Default to PAGE_SIZE after much discussion.
631 */
632
633 if (vap->va_type == VREG) {
634 sb->st_blksize = vap->va_blocksize;
635 } else if (vn_isdisk(vp, NULL)) {
636 sb->st_blksize = vp->v_rdev->si_bsize_best;
637 if (sb->st_blksize < vp->v_rdev->si_bsize_phys)
638 sb->st_blksize = vp->v_rdev->si_bsize_phys;
639 if (sb->st_blksize < BLKDEV_IOSIZE)
640 sb->st_blksize = BLKDEV_IOSIZE;
641 } else {
642 sb->st_blksize = PAGE_SIZE;
643 }
644
645 sb->st_flags = vap->va_flags;
646 if (suser(td))
647 sb->st_gen = 0;
648 else
649 sb->st_gen = vap->va_gen;
650
651#if (S_BLKSIZE == 512)
652 /* Optimize this case */
653 sb->st_blocks = vap->va_bytes >> 9;
654#else
655 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
656#endif
657 return (0);
658}
659
660/*
661 * File table vnode ioctl routine.
662 */
663static int
664vn_ioctl(fp, com, data, td)
665 struct file *fp;
666 u_long com;
667 void *data;
668 struct thread *td;
669{
670 register struct vnode *vp = ((struct vnode *)fp->f_data);
671 struct vnode *vpold;
672 struct vattr vattr;
673 int error;
674
675 switch (vp->v_type) {
676
677 case VREG:
678 case VDIR:
679 if (com == FIONREAD) {
680 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
681 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
682 VOP_UNLOCK(vp, 0, td);
683 if (error)
684 return (error);
685 *(int *)data = vattr.va_size - fp->f_offset;
686 return (0);
687 }
688 if (com == FIONBIO || com == FIOASYNC) /* XXX */
689 return (0); /* XXX */
690 /* fall into ... */
691
692 default:
693#if 0
694 return (ENOTTY);
695#endif
696 case VFIFO:
697 case VCHR:
698 case VBLK:
699 if (com == FIODTYPE) {
700 if (vp->v_type != VCHR && vp->v_type != VBLK)
701 return (ENOTTY);
702 *(int *)data = devsw(vp->v_rdev)->d_flags & D_TYPEMASK;
703 return (0);
704 }
705 error = VOP_IOCTL(vp, com, data, fp->f_flag, td->td_ucred, td);
706 if (error == 0 && com == TIOCSCTTY) {
707
708 /* Do nothing if reassigning same control tty */
709 sx_slock(&proctree_lock);
710 if (td->td_proc->p_session->s_ttyvp == vp) {
711 sx_sunlock(&proctree_lock);
712 return (0);
713 }
714
715 vpold = td->td_proc->p_session->s_ttyvp;
716 VREF(vp);
717 SESS_LOCK(td->td_proc->p_session);
718 td->td_proc->p_session->s_ttyvp = vp;
719 SESS_UNLOCK(td->td_proc->p_session);
720
721 sx_sunlock(&proctree_lock);
722
723 /* Get rid of reference to old control tty */
724 if (vpold)
725 vrele(vpold);
726 }
727 return (error);
728 }
729}
730
731/*
732 * File table vnode poll routine.
733 */
734static int
735vn_poll(fp, events, cred, td)
736 struct file *fp;
737 int events;
738 struct ucred *cred;
739 struct thread *td;
740{
741
742 return (VOP_POLL(((struct vnode *)fp->f_data), events, cred, td));
743}
744
745/*
746 * Check that the vnode is still valid, and if so
747 * acquire requested lock.
748 */
749int
750#ifndef DEBUG_LOCKS
751vn_lock(vp, flags, td)
752#else
753debug_vn_lock(vp, flags, td, filename, line)
754#endif
755 struct vnode *vp;
756 int flags;
757 struct thread *td;
758#ifdef DEBUG_LOCKS
759 const char *filename;
760 int line;
761#endif
762{
763 int error;
764
765 do {
766 if ((flags & LK_INTERLOCK) == 0)
767 mtx_lock(&vp->v_interlock);
768 if ((vp->v_flag & VXLOCK) && vp->v_vxproc != curthread) {
769 vp->v_flag |= VXWANT;
770 msleep(vp, &vp->v_interlock, PINOD | PDROP,
771 "vn_lock", 0);
772 error = ENOENT;
773 } else {
774#if 0
775 /* this can now occur in normal operation */
776 if (vp->v_vxproc != NULL)
777 log(LOG_INFO, "VXLOCK interlock avoided in vn_lock\n");
778#endif
779#ifdef DEBUG_LOCKS
780 vp->filename = filename;
781 vp->line = line;
782#endif
783 error = VOP_LOCK(vp,
784 flags | LK_NOPAUSE | LK_INTERLOCK, td);
785 if (error == 0)
786 return (error);
787 }
788 flags &= ~LK_INTERLOCK;
789 } while (flags & LK_RETRY);
790 return (error);
791}
792
793/*
794 * File table vnode close routine.
795 */
796static int
797vn_closefile(fp, td)
798 struct file *fp;
799 struct thread *td;
800{
801
802 fp->f_ops = &badfileops;
803 return (vn_close(((struct vnode *)fp->f_data), fp->f_flag,
804 fp->f_cred, td));
805}
806
807/*
808 * Preparing to start a filesystem write operation. If the operation is
809 * permitted, then we bump the count of operations in progress and
810 * proceed. If a suspend request is in progress, we wait until the
811 * suspension is over, and then proceed.
812 */
813int
814vn_start_write(vp, mpp, flags)
815 struct vnode *vp;
816 struct mount **mpp;
817 int flags;
818{
819 struct mount *mp;
820 int error;
821
822 /*
823 * If a vnode is provided, get and return the mount point that
824 * to which it will write.
825 */
826 if (vp != NULL) {
827 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
828 *mpp = NULL;
829 if (error != EOPNOTSUPP)
830 return (error);
831 return (0);
832 }
833 }
834 if ((mp = *mpp) == NULL)
835 return (0);
836 /*
837 * Check on status of suspension.
838 */
839 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
840 if (flags & V_NOWAIT)
841 return (EWOULDBLOCK);
842 error = tsleep(&mp->mnt_flag, (PUSER - 1) | (flags & PCATCH),
843 "suspfs", 0);
844 if (error)
845 return (error);
846 }
847 if (flags & V_XSLEEP)
848 return (0);
849 mp->mnt_writeopcount++;
850 return (0);
851}
852
853/*
854 * Secondary suspension. Used by operations such as vop_inactive
855 * routines that are needed by the higher level functions. These
856 * are allowed to proceed until all the higher level functions have
857 * completed (indicated by mnt_writeopcount dropping to zero). At that
858 * time, these operations are halted until the suspension is over.
859 */
860int
861vn_write_suspend_wait(vp, mp, flags)
862 struct vnode *vp;
863 struct mount *mp;
864 int flags;
865{
866 int error;
867
868 if (vp != NULL) {
869 if ((error = VOP_GETWRITEMOUNT(vp, &mp)) != 0) {
870 if (error != EOPNOTSUPP)
871 return (error);
872 return (0);
873 }
874 }
875 /*
876 * If we are not suspended or have not yet reached suspended
877 * mode, then let the operation proceed.
878 */
879 if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPENDED) == 0)
880 return (0);
881 if (flags & V_NOWAIT)
882 return (EWOULDBLOCK);
883 /*
884 * Wait for the suspension to finish.
885 */
886 return (tsleep(&mp->mnt_flag, (PUSER - 1) | (flags & PCATCH),
887 "suspfs", 0));
888}
889
890/*
891 * Filesystem write operation has completed. If we are suspending and this
892 * operation is the last one, notify the suspender that the suspension is
893 * now in effect.
894 */
895void
896vn_finished_write(mp)
897 struct mount *mp;
898{
899
900 if (mp == NULL)
901 return;
902 mp->mnt_writeopcount--;
903 if (mp->mnt_writeopcount < 0)
904 panic("vn_finished_write: neg cnt");
905 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
906 mp->mnt_writeopcount <= 0)
907 wakeup(&mp->mnt_writeopcount);
908}
909
910/*
911 * Request a filesystem to suspend write operations.
912 */
913void
914vfs_write_suspend(mp)
915 struct mount *mp;
916{
917 struct thread *td = curthread;
918
919 if (mp->mnt_kern_flag & MNTK_SUSPEND)
920 return;
921 mp->mnt_kern_flag |= MNTK_SUSPEND;
922 if (mp->mnt_writeopcount > 0)
923 (void) tsleep(&mp->mnt_writeopcount, PUSER - 1, "suspwt", 0);
924 VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td);
925 mp->mnt_kern_flag |= MNTK_SUSPENDED;
926}
927
928/*
929 * Request a filesystem to resume write operations.
930 */
931void
932vfs_write_resume(mp)
933 struct mount *mp;
934{
935
936 if ((mp->mnt_kern_flag & MNTK_SUSPEND) == 0)
937 return;
938 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPENDED);
939 wakeup(&mp->mnt_writeopcount);
940 wakeup(&mp->mnt_flag);
941}
942
943/*
944 * Implement kqueues for files by translating it to vnode operation.
945 */
946static int
947vn_kqfilter(struct file *fp, struct knote *kn)
948{
949
950 return (VOP_KQFILTER(((struct vnode *)fp->f_data), kn));
951}
952
953/*
954 * Simplified in-kernel wrapper calls for extended attribute access.
955 * Both calls pass in a NULL credential, authorizing as "kernel" access.
956 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
957 */
958int
959vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
960 const char *attrname, int *buflen, char *buf, struct thread *td)
961{
962 struct uio auio;
963 struct iovec iov;
964 int error;
965
966 iov.iov_len = *buflen;
967 iov.iov_base = buf;
968
969 auio.uio_iov = &iov;
970 auio.uio_iovcnt = 1;
971 auio.uio_rw = UIO_READ;
972 auio.uio_segflg = UIO_SYSSPACE;
973 auio.uio_td = td;
974 auio.uio_offset = 0;
975 auio.uio_resid = *buflen;
976
977 if ((ioflg & IO_NODELOCKED) == 0)
978 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
979
980 /* authorize attribute retrieval as kernel */
981 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
982 td);
983
984 if ((ioflg & IO_NODELOCKED) == 0)
985 VOP_UNLOCK(vp, 0, td);
986
987 if (error == 0) {
988 *buflen = *buflen - auio.uio_resid;
989 }
990
991 return (error);
992}
993
994/*
995 * XXX failure mode if partially written?
996 */
997int
998vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
999 const char *attrname, int buflen, char *buf, struct thread *td)
1000{
1001 struct uio auio;
1002 struct iovec iov;
1003 struct mount *mp;
1004 int error;
1005
1006 iov.iov_len = buflen;
1007 iov.iov_base = buf;
1008
1009 auio.uio_iov = &iov;
1010 auio.uio_iovcnt = 1;
1011 auio.uio_rw = UIO_WRITE;
1012 auio.uio_segflg = UIO_SYSSPACE;
1013 auio.uio_td = td;
1014 auio.uio_offset = 0;
1015 auio.uio_resid = buflen;
1016
1017 if ((ioflg & IO_NODELOCKED) == 0) {
1018 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1019 return (error);
1020 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1021 }
1022
1023 /* authorize attribute setting as kernel */
1024 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1025
1026 if ((ioflg & IO_NODELOCKED) == 0) {
1027 vn_finished_write(mp);
1028 VOP_UNLOCK(vp, 0, td);
1029 }
1030
1031 return (error);
1032}
1033
1034int
1035vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1036 const char *attrname, struct thread *td)
1037{
1038 struct mount *mp;
1039 int error;
1040
1041 if ((ioflg & IO_NODELOCKED) == 0) {
1042 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1043 return (error);
1044 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1045 }
1046
1047 /* authorize attribute removal as kernel */
1048 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td);
1049
1050 if ((ioflg & IO_NODELOCKED) == 0) {
1051 vn_finished_write(mp);
1052 VOP_UNLOCK(vp, 0, td);
1053 }
1054
1055 return (error);
1056}